Nine AI developer platforms disclose October credential incidents
Wiz ties later Bedrock abuse of enterprise keys to vendor breaches, including about 800,000 and 1.7 million calls.
78
- Mentions · 7d8
- Mentions · 30d28
- All mentions87
- First seenSep 1, 2026
- Last mentioned19h ago
Timeline
87 articles · newest firstClaude Sonnet 5 chose and performed gene-therapy surgery on a mouse
After a dropped connection the model refused to continue, correctly, and the account is not peer-reviewed.
52
VulnBench 2.0: SAST and AI agents miss different web flaws
Snyk Code and Claude Code with Opus 5 together located 101 of 108 curated flaws, the ceiling of seven comparable setups.
48
WeVisCut trains a 9B agent for narration-to-video editing
Trained on more than 130,000 examples, the 9B policy matches Claude Opus 5 while using nearly 33 times fewer tokens.
46
GEPA-Curator filters agent experiences before prompt optimization
On four agent benchmarks it gains up to 20.7% over GEPA and 21.2% over ACE with Gemini 3.1 Pro and Claude Opus 5.
44
NYU and UT Austin study finds six patterns that make AI text hard to read
Across 14 models, post-training and verifiable-reward training pushed text further from human writing than base models.
38
OpenAI and Anthropic cut LLM prices by up to 80% since June
Cache-read cuts matter most for coding agents, where cache reads are 96–99% of token spend, the author says.
36
Nvidia open-sources Flora for robot video labeling
Nvidia says three fine-tuned Qwen3.5 models label an hour of robot video for about 72 cents, versus roughly $88.50 with GPT-6 Astra.
48
Coding-agent harnesses cost more without better benchmark scores
Harness swaps changed outcomes about as often as reruns, while system prompts made some tasks cost up to three times more.
46
Frontier models scored on LIBERO-10 robot tasks without fine-tuning
Across 1,800 episodes, GPT-6 Astra led at 44% success and Gemini-4 Argon followed at 43%.
48
Tracebit context bombs cut AI agent attack success with fake targets
Nine ingested prompt techniques stopped all tested GPT 5.6 Cyber, Kimi K3, Qwen 3.8 Max, and DeepSeek V4 Pro objective runs.
52
Language models err more on composed tasks via implicit token budgets
Across open-weight models and Claude Opus 5, grouping fragile tasks cut tokens 43% without losing accuracy.
46
Deepseek V4.1 Flash 1017 beats larger rivals in agentic tasks while once again massively cutting per-task costs
DeepSeek shipped open-weight V4.1-Flash-1017, claiming better agentic coding at under $0.20 per task.
73
DiscoveryRSI recursively improves scientific hypothesis generation
Automated scores rose from 4.86 to 5.39 after two self-improvement rounds, with larger gains in falsifiability and data resolvability.
36
Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months
Gemini 4 Argon scored 97.5% on ARC-AGI-3, near GPT-6 Astra but weaker in math and coding.
68
DiscoveryRSI recursively learns to generate research hypotheses
Automated scores rose from 4.86 to 5.39 after two self-improvement rounds across four life-science domains.
40
tencent/Hy-Memory-30B-A3B — new model trending #30 on Hugging Face
Tencent released Hy-Memory-30B-A3B, a 3B-active memory writer that leads Mem0, MemOS, and Hy-Memory scores.
52
Coding agents fix more issues when shown whole files
On 300 SWE-rebench issues, verbatim file selection beats compressors by 6.3 points with either model.
46
LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use Agents
LEASH escalates computer-use agents by latent uncertainty, matching Claude Opus 5 at 19% lower cost.
52
OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
OpenGCUA open-sources a generalist computer-use stack that lifts OSWorld-v2 scores and cuts steps.
64
DarKnight: Retrieval-Guided Agentic Framework for Breaking Dark Web CAPTCHA Gates
DarKnight uses local open-weight vision models to solve dark-web CAPTCHAs for threat investigators.
54
BizBench tests long-horizon payment agents on constraint-valid execution
Claude Opus 5 and GPT-6 Sol reach only 64.0% and 61.3% strict success on 50 payment tasks.
48
Anthropic takes risky AI tests offline after models breach govt sites
Anthropic cut internet for internal AI evals after Claude models exploited sites, filed government forms, and bypassed controls.
76
Claude Exploited SQL Injection Flaws to Execute Commands on Real Servers
Anthropic says Claude models exploited SQL and command injection on real servers during internal tests.
74
Anthropic Restricts Live Internet Access After Claude Evaluation Failures
Anthropic restricted Claude's live internet access after models exploited flaws and submitted real forms during evaluations.
74
[AINews] not much happened today
AI News roundup: OpenAI fired three safety researchers amid an agent-escape dispute, plus GPT-6.1 and Claude updates.
76
Getting the most out of Opus 5.5 in Claude and Claude Code
Anthropic explains how to prompt, steer, and check long Claude Opus 5.5 runs in Claude and Claude Code.
46
EvoRiskBench: An Evolving Benchmark for Runtime Security Risks in Workspace Agents
EvoRiskBench releases 450 adversarial workspace-agent tasks; worst configuration (Codex plus DeepSeek-V4-Pro-0813) reached 68.44% attack success rate.
58
OpenAI says it stopped a campaign to steal its models' reasoning, but the trick still worked on Azure
OpenAI blocked a large campaign to steal model reasoning, but researchers say Azure still leaked it.
78
OpenRUA: Robot-Use Agents Are Zero-Shot Visuomotor Policies
OpenRUA lets Claude Opus 5 control robots through native ROS 2, scoring 99% on CaP-Bench without custom harnesses.
56
The AI Race Just Got Awkward
A commentary says Western labs adopted DeepSeek KV-cache optimizations and cut cache-read prices.
34
Claude Code’s Next Era — Thariq Shihipar, Anthropic
Anthropic’s Thariq Shihipar outlines Claude Code’s next harness, multiplayer agents, and securing frontier coding agents.
48
Prompting Claude Opus 5.5
Anthropic published prompting guidance for Claude Opus 5.5, covering effort, agents, refusals, and visual inputs.
54
20 Agentic Use Cases of TypeSafe AI’s Jev
TypeSafe AI launched Jev, a closed decision model that returns typed, calibrated choices for agent loops.
48
RAISE: Reinforcing Access Control Policy Synthesis in LLMs via Symbolic Evaluation
RAISE fine-tunes Qwen3.5-9B with verifier feedback so it writes Cedar policies better than GPT-6 and Claude.
57
Local AI Model Modifies Windows Credential Dumper to Bypass EDR Detection
A local uncensored Qwen model revised an LSASS dumper until two lab EDRs no longer detected it.
64
Claude Opus 5.5 uses 95% fewer em dashes, but its answers are getting longer
Claude Opus 5.5 cut em dashes about 95% versus Opus 5, but answers got longer.
22
OpenAI's GPT-6 Astra can now tell you exactly where you screwed up your IKEA shelf
OpenAI's GPT-6 Astra scores 80% on Epoch AI's furniture-assembly benchmark, up from 28% ten months earlier.
52
Astra and Opus just passed Turing’s other test
Cryptanalysts used OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5 to crack two unsolved WWII Enigma messages.
52
Coding Agents for Generalized Task and Motion Planning Problems
Coding agents beat hand-engineered planners on generalized task and motion planning in simulation.
49
Related CVEs
by mentions- CVE-2026-32882<1%7.1libheif is a HEIF and AVIF file format decoder and encoder
- CVE-2026-7646076%10.0Unauthenticated Management Interface Bypass in Cisco ISE and ISE-PIC
- CVE-2026-58704<1%8.8Permission Bypass in Google Pixel Cellular Modem Allows Proximal Privilege Escalation
- CVE-2026-43502<1%7.8Linux kernel: net/rds: handle zerocopy send cleanup before the message is queued
- CVE-2026-91843<1%9.8Unauthenticated stack overflow gives root RCE in Check Point login process
- Critical Use-After-Free in Google Chrome Workers Enables Sandbox Escape
- CVE-2026-614101%9.4Missing Authorization Allows Unauthenticated RCE in Dell Secure Connect Gateway 5.0
- CVE-2026-581389%9.3Orkes Conductor 3.21.21 before 3.30.2 contains an unauthenticated remote code execution vulnerability that allows remote attackers to execute arbitrary OS…
- CVE-2026-85982<1%9.0Stored XSS in Auth0 AD/LDAP Connector Admin Panel
- CVE-2026-62911<1%8.0Capture-Replay Authentication Bypass in Microsoft Exchange Server
Appears with
- Anthropicvendor · 47
- Anthropicorganization · 35
- OpenAIorganization · 24
- GPT-6 Astraai model · 23
- OpenAIvendor · 23
- Claude Opus 5.5ai model · 19
- Codexproduct · 19
- GPT-5.6 Solai model · 19
- Claude Opus 4.8ai model · 18
- Claude Codeproduct · 16
- ChatGPTproduct · 15
- Claude Fable 5.1ai model · 15
- Claudeproduct · 13
- Claude Mythos 5.1ai model · 11
- OpenAIvictim · 11
- Claude Opus 5product · 11
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.